SGE2010P & SGE2000P LAG configuration

[The_Uber_CaT]

Hi,

I've got multiple SGE2000P & SGE2010P switches

Originally I was happy just set them all up in a stacked ring configuration.

However I've become aware that my back bone fibre links are contantly pushed to their max.

So I've decided that I should change this configuration and unstack the switches and make them all stand-alone units.

and I'll configure 2 cable LAG links between all my switches Theorically I should now get 2GIGs between each switch

and to complete the ring in my network for redundancy I'll turn on Spanning tree.

However I've tested the speed and I just can not seem to get a LAG connection with 2 x 1gig cables to push more data than a single link 1gig link cable would.

Why is this?

My test enviroment(not using fibre optic cables just cat5e copper cabling):

4 x pc's(all gigabit network cards)

2 x sge2000p switches

PC1 ---(1G eth)---\                                                                        /---(1G eth)--- PC2

                       [SGE2000P]===(LAG1 2x1G cables)===[SGE2000P]

PC3 ---(1G eth)---/                                                                        \---(1G eth)--- PC4

If I send files from PC1 & PC3 simultaneously to PC2 & PC4

They don't transfer faster than if I I just use a single 1GIG Link cable

Please can someone tell me what I'm doing wrong?

looking at the LAG configuration it shows both cables are connected & the LAG looks like it's working.

But it really doesn't seem to be running at the expected 2GIG?

The LAG fail over seems to work fine if I remove either of the 2 cables from the LAG the link continues to work.

(sometimes it will drop a ping when removing or readding a LAG cable)

Thanks

Cheers

-Matt

[David Hornstein]

Hi Matt,

Check out my old posting on LAG and see if your observed behaviour now makes sense,.

https://supportforums.cisco.com/thread/2101531

regards Dave

[The_Uber_CaT]

Ok so I've read your artical,

and if I understand correctly no single data stream between any 2 given PC's will go faster than 1GIG

(even if your LAG is 2x1gigcables = 2gig)

and that is what I assumed anyway.

But what I'm saying is

when I similtaniously transfer data from PC1 to PC2 & transfer different data from PC3 to PC4

I both transfers should be faster if I've got a LAG connection running between the switches.

But when I start the 2nd transfer between 2 different PC's the speed on the first 2 PC's gets cut in half.

as if the LAG connection is not load balancing.

in fact I tried pushing it futher I added another 2 computers and started a 3rd data transfer between those 2 PC's

and my transfer rates all dropped to a 3rd of what they were doing.

I know the LAG is working because I can remove either of the 2 LAG cables and the link continues to work

it just doesn't seem to do any load balancing it's as if its only ever using 1 of the 2 gigabit cables from the LAG at a time.

in fact if I watch the activity lights on the 2 LAG cables while transfering data only 1 of the 2 lights really flash like mad and if I pull it out then the activity swaps over to the other cable.

surely if I've got 3 seperate data copy streams running between 6 different computers it SHOULD balance that transfer to some degree accross the 2 gigabit LAG cables?

Thanks

Cheers

-Matt

[David Hornstein]

Hi Matt,

Yep, I understand your aspirations as to how you expect LAG to work.

But the tiny brain inside the switch uses a algorithm to determine which link in a LAG to send the traffic over. This determination is set according to rules set out in the standard.

Yes, with more hosts, it will spread the traffic more evenly, but with such few hosts in your test, statistically your just out of luck that the switch looked at the source ip/Mac and destination ip/Mac and determined to  sent the traffic over the same link.

With such few hosts, as you have already seen, LAG in your case has only provided link redundancy.

With a lot more IP hosts you will notice a more even spread of traffic. Statistically, at this stage of your testing, your just out of  luck.  Try it in a live network, and then check (after a few hours) via looking at the port counters, you should hopefully notice a more equitable spread of bytes between the lag ports.

Regards Dave